I am currently building a GPU compute system, and can find dual Socket 2011 motherboards with 8 PCIE 3.0x16 double-width slots. I will be filling these with orphaned AMD Radeon HD 7990 GPUs. My question is, will getting 2 cheap Xeons bottleneck these in simulation and CAD applications like ANSYS and CATIA?
Thanks!
:D
If you want to run 8 7990s, 4 on each CPU, you are going to need at least 2 8 core Xeon Processors or 2 16 core Opteron processors. The CPU to GPU relation between that much GPU power is insane. Also you will not get benefit from 4 7990s as there is a software limit to 4 GPUs per crossfire configuration. This of course being null if we were talking about GPU mining.
For the applications he's using the computer for he will benefit from more than 4 cards. Cad and those applications will use all those graphics cards.
Even if that is the case, a pair of 4 core Xeons aren't going to be enough to deal with the raw amount of GPU data coming from 16 GPU cores. 8 Xeon cores minimum, 16 Opteron cores minimum per cpu pair on a dual socket board.
Thank you :D
You have helped me capture the elusive Potogan.
When it comes to incoming data, will there be enough PCI-E bandwidth for the GPUs? I understand that not all of them will have a full PCI-E 3.0 x 16 connection. So, will PCI-E 3.0 x 8 slots or multiplexed x16 slots have a significant effect on performance? (Will the GPUs saturate the PCI-E bandwidth provided by the dual Xeons?)
Do you have any personal experience with any of these programs? (Their local forums are, simply put, dead. No one goes there anymore to answer questions).